Connection pool: trigger reconnection on fatal errors

If a pooled connection encounters a fatal error (as per #253), we should trigger a re-connection. Currently:
* If the connection gets reset (the default), the error gets detected and re-connection triggers - but we're wasting a server round-trip.
* If the connection doesn't get reset (as per `return_without_reset`), the connection will never be reconnected, which can be problematic.

Start by reading issue #253 to understand the defined fatal-error behavior, then locate the connection-pool handling for reset and return_without_reset cases. Done means a fatal error triggers reconnection without the unnecessary server round-trip and also reconnects connections returned without reset; no files or tests are named in the issue.

Written by the indexing model from the issue text.